1
00:00:00,080 --> 00:00:03,800
Speaker 1: The conversation is changing right from an AI wave driven

2
00:00:03,920 --> 00:00:07,800
Speaker 1: by startups like open ai to sovereign AI. And you

3
00:00:07,800 --> 00:00:11,559
Speaker 1: yourself are in Toronto, Canada right now. My understanding is

4
00:00:11,640 --> 00:00:14,760
Speaker 1: you've just met with a Canadian government official. What did

5
00:00:14,800 --> 00:00:15,280
Speaker 1: you agree on?

6
00:00:18,640 --> 00:00:22,320
Speaker 2: I met with Minister Champaign and I'm here. I'm here

7
00:00:22,320 --> 00:00:27,160
Speaker 2: for several reasons. One to celebrate the research center that

8
00:00:27,200 --> 00:00:30,720
Speaker 2: we have here in Toronto. You probably know that the

9
00:00:30,880 --> 00:00:34,800
Speaker 2: University of Toronto is as close to the epicenter of

10
00:00:34,840 --> 00:00:38,440
Speaker 2: the invention of modern AI as it gets. Professor Hin

11
00:00:38,720 --> 00:00:43,640
Speaker 2: was here the advances in deep learning. Alias Suskober was

12
00:00:43,640 --> 00:00:45,840
Speaker 2: a student, and so I had met many of them

13
00:00:46,320 --> 00:00:50,520
Speaker 2: in the early days of the invention of modern AI,

14
00:00:51,159 --> 00:00:54,920
Speaker 2: and so we started our research lab here. There's a

15
00:00:55,000 --> 00:01:00,360
Speaker 2: large AI community here, and what is needed is for

16
00:01:00,480 --> 00:01:04,720
Speaker 2: Canada to have a public infrastructure that supports the ongoing

17
00:01:04,760 --> 00:01:08,839
Speaker 2: advancement of AI research and to support the local startups

18
00:01:08,880 --> 00:01:12,319
Speaker 2: and the local industries. One of the things about generative

19
00:01:12,319 --> 00:01:15,479
Speaker 2: AI is is a form of computer science, a form

20
00:01:15,520 --> 00:01:20,560
Speaker 2: of computing where a large computing infrastructure is essential to

21
00:01:21,000 --> 00:01:25,320
Speaker 2: the creation of the model, the large language model, as

22
00:01:25,400 --> 00:01:29,280
Speaker 2: well as the generation the tokens and the information that

23
00:01:29,400 --> 00:01:33,280
Speaker 2: is really valuable that comes out of generative AI. And

24
00:01:32,680 --> 00:01:36,720
Speaker 2: so most of our conversation is around the AI infrastructure

25
00:01:36,840 --> 00:01:40,200
Speaker 2: necessary in Canada so that Canada could have its own

26
00:01:40,280 --> 00:01:41,760
Speaker 2: sovereign AI capabilities.

27
00:01:42,800 --> 00:01:47,360
Speaker 1: You've spoken Jensen about the nation state level GPU specialized

28
00:01:47,360 --> 00:01:50,880
Speaker 1: cloud or sovereign AI specialized cloud. And then we look

29
00:01:50,920 --> 00:01:53,320
Speaker 1: at what the hyper scalers are doing. You know, you've

30
00:01:53,360 --> 00:01:55,800
Speaker 1: reflected on that. When we talk about H one hundred,

31
00:01:56,120 --> 00:01:59,320
Speaker 1: we're really talking about HGX server design, right or DGX

32
00:01:59,360 --> 00:02:02,680
Speaker 1: server design, and that ships to the hyperscalers. So what

33
00:02:02,720 --> 00:02:06,800
Speaker 1: does this business line where you deal directly with governments

34
00:02:06,840 --> 00:02:09,960
Speaker 1: in nation states add to what those hyper scalers are

35
00:02:10,000 --> 00:02:11,640
Speaker 1: already doing in the private sector?

36
00:02:13,639 --> 00:02:17,280
Speaker 2: Well as much as as much as possible, we should,

37
00:02:17,520 --> 00:02:23,519
Speaker 2: We should utilize, utilize clouds and it's accessible, the technology

38
00:02:23,560 --> 00:02:27,320
Speaker 2: is modern. Our partnership with Microsoft and a w S

39
00:02:27,440 --> 00:02:33,359
Speaker 2: and gcp UH and the other cloud service providers make

40
00:02:33,400 --> 00:02:36,560
Speaker 2: it make it, make the technology accessible so that everybody

41
00:02:36,560 --> 00:02:40,280
Speaker 2: could could have benefits from it. There are also regional

42
00:02:40,320 --> 00:02:43,920
Speaker 2: capabilities in each one of the countries that would like

43
00:02:44,040 --> 00:02:48,400
Speaker 2: to be able to build on this important infrastructure. Artificial

44
00:02:48,440 --> 00:02:54,520
Speaker 2: intelligence is about the generation of the generation of tokens

45
00:02:54,520 --> 00:02:56,600
Speaker 2: and the final analysis, these are numbers that come out

46
00:02:56,600 --> 00:03:00,640
Speaker 2: of computers that is interpreted as intelligence in each one

47
00:03:00,639 --> 00:03:03,280
Speaker 2: of the various domains that we serve. And so the

48
00:03:04,280 --> 00:03:08,200
Speaker 2: production of these tokens, the generation of these tokens require

49
00:03:08,280 --> 00:03:13,160
Speaker 2: AI supercomputers, and so we have the we have the technology,

50
00:03:13,200 --> 00:03:17,679
Speaker 2: the know how, the full stack from the compute networking

51
00:03:18,360 --> 00:03:21,520
Speaker 2: to the software stack to enable each one of the

52
00:03:21,840 --> 00:03:23,440
Speaker 2: each one of the countries to be able to build

53
00:03:23,440 --> 00:03:27,720
Speaker 2: their own AI infrastructure. And so we're I think you're

54
00:03:27,720 --> 00:03:30,520
Speaker 2: going to see countries around the world be able to

55
00:03:30,960 --> 00:03:35,040
Speaker 2: continue to use public clouds, but also build regional regional

56
00:03:35,160 --> 00:03:40,000
Speaker 2: data centers as well as publicly supported infrastructure so that

57
00:03:40,080 --> 00:03:42,120
Speaker 2: each one of the countries could be able to cultivate

58
00:03:42,360 --> 00:03:43,840
Speaker 2: and advance its own industries.

59
00:03:45,120 --> 00:03:48,280
Speaker 1: Jensen to the governments that you speak to understand those

60
00:03:48,360 --> 00:03:52,559
Speaker 1: dynamics with respect. Politics tends to move a lot slower

61
00:03:52,600 --> 00:03:56,040
Speaker 1: than the rate of technological progress we've seen in just

62
00:03:56,080 --> 00:03:57,040
Speaker 1: the last twelve months.

63
00:03:57,040 --> 00:04:06,160
Speaker 2: For example, the last twelve months, you you have seen UH, India, Japan,

64
00:04:06,400 --> 00:04:13,680
Speaker 2: and France, Canada, now Southeast Asia, Singapore, UH speak up

65
00:04:13,800 --> 00:04:18,600
Speaker 2: about the importance of investing in sovereign AI capabilities. Uh,

66
00:04:18,920 --> 00:04:21,880
Speaker 2: it is become abundantly clear to each one of the

67
00:04:21,880 --> 00:04:25,920
Speaker 2: countries that that their natural resource, which is the data

68
00:04:25,960 --> 00:04:31,080
Speaker 2: of their country, should be should be refined and produced

69
00:04:31,320 --> 00:04:35,760
Speaker 2: intelligence of their country for their country. And that capability

70
00:04:36,320 --> 00:04:38,880
Speaker 2: of refining the data of their country, of their country

71
00:04:38,920 --> 00:04:42,880
Speaker 2: and turn it into their artificial intelligence is now possible

72
00:04:42,920 --> 00:04:46,160
Speaker 2: in a quite a quite a democratized way. Almost every

73
00:04:46,200 --> 00:04:49,160
Speaker 2: country should be able to do it for themselves. And

74
00:04:49,160 --> 00:04:53,120
Speaker 2: and what's needed, of course is the technology and the

75
00:04:53,200 --> 00:04:56,920
Speaker 2: know how of standing up AI infrastructure, and that's where

76
00:04:56,960 --> 00:05:01,240
Speaker 2: we could be quite helpful to to various regions. And

77
00:05:01,279 --> 00:05:04,880
Speaker 2: so I think that the recognition of the importance of

78
00:05:04,880 --> 00:05:08,320
Speaker 2: sovereign AI capabilities is now quite global.

79
00:05:09,400 --> 00:05:12,840
Speaker 1: Jensen, does that recognition and your ability to help extend

80
00:05:12,880 --> 00:05:15,880
Speaker 1: to China? You know on my own show Bloomberg Technology,

81
00:05:15,920 --> 00:05:20,400
Speaker 1: the academic community and the startup community reflect that there

82
00:05:20,480 --> 00:05:23,440
Speaker 1: is a desire at the nation state level to have

83
00:05:23,720 --> 00:05:27,560
Speaker 1: sovereign AI competency. But there's also a lot of work

84
00:05:27,600 --> 00:05:30,920
Speaker 1: going on with companies by DO as an example. Are

85
00:05:30,960 --> 00:05:33,480
Speaker 1: you confident that you will be able to work with

86
00:05:33,560 --> 00:05:37,040
Speaker 1: China on the topic of sovereign AI going forward, given

87
00:05:37,440 --> 00:05:39,200
Speaker 1: the political backdrop that we live in.

88
00:05:41,400 --> 00:05:44,080
Speaker 2: What we're American company and we have to comply with

89
00:05:44,279 --> 00:05:48,279
Speaker 2: American policies and whatever the rules and regulations are and

90
00:05:48,320 --> 00:05:51,920
Speaker 2: the laws are, will number one, comply with that, work

91
00:05:51,960 --> 00:05:57,840
Speaker 2: closely with regulators and understand understand their intentions and their desires,

92
00:05:58,560 --> 00:06:01,960
Speaker 2: work within those boundaries UH and be able to create

93
00:06:02,000 --> 00:06:06,839
Speaker 2: products for for the various countries that are involved, fully

94
00:06:06,839 --> 00:06:10,160
Speaker 2: in compliant with the regulations that that are that are

95
00:06:10,160 --> 00:06:13,520
Speaker 2: in front of us, UH and beyond that once we

96
00:06:13,680 --> 00:06:16,920
Speaker 2: once we comply, our goal are in the United States

97
00:06:16,960 --> 00:06:19,599
Speaker 2: would love to see us be a successful country and

98
00:06:19,600 --> 00:06:23,040
Speaker 2: in one of the pillars of national security of successful

99
00:06:23,080 --> 00:06:28,200
Speaker 2: industries and it creates jobs UH and UH allows our

100
00:06:28,240 --> 00:06:33,320
Speaker 2: country to stay ahead and technologically and so UH it

101
00:06:33,400 --> 00:06:36,680
Speaker 2: is of a great interest of our nation that our

102
00:06:36,800 --> 00:06:39,960
Speaker 2: American companies are successful around the world. And so once

103
00:06:40,160 --> 00:06:42,839
Speaker 2: once we comply with the regulations, we'll do our best

104
00:06:42,880 --> 00:06:46,880
Speaker 2: to serve the local markets. And UH we have full

105
00:06:47,160 --> 00:06:52,680
Speaker 2: the We we have excellent communications with with the administration,

106
00:06:52,880 --> 00:06:56,720
Speaker 2: and we have UH in working in full compliance being

107
00:06:56,720 --> 00:06:58,560
Speaker 2: able to serve the local markets. And we have to

108
00:06:58,680 --> 00:07:00,120
Speaker 2: force support.

109
00:07:00,680 --> 00:07:04,000
Speaker 1: Agen so I go back to that HGX or DGX example. Right,

110
00:07:04,040 --> 00:07:07,440
Speaker 1: we understand how it works with private companies and cloud

111
00:07:07,920 --> 00:07:12,720
Speaker 1: going forward, how should we think about sovereign AI as

112
00:07:12,760 --> 00:07:15,280
Speaker 1: a business line for you? Is there a way that

113
00:07:15,360 --> 00:07:18,920
Speaker 1: we can understand how n videos work even if it's

114
00:07:18,920 --> 00:07:22,960
Speaker 1: building supercomputers like in the UK, for example, what proportion

115
00:07:23,120 --> 00:07:26,640
Speaker 1: of your overall business that will represent if countries is

116
00:07:26,640 --> 00:07:30,480
Speaker 1: to lead the way, the.

117
00:07:30,520 --> 00:07:35,400
Speaker 2: Vast majority of the computing market has been the United

118
00:07:35,440 --> 00:07:41,040
Speaker 2: States and to a small to a much longer smaller degree, China.

119
00:07:41,120 --> 00:07:45,280
Speaker 2: For the very first time, every industry would be every

120
00:07:45,320 --> 00:07:49,040
Speaker 2: single country will become a computer industry, and every industry

121
00:07:49,080 --> 00:07:53,480
Speaker 2: will become a technology industry. And so artificial intelligence or

122
00:07:53,520 --> 00:07:59,280
Speaker 2: the automation, the production at scale of intelligence matters to

123
00:07:59,320 --> 00:08:01,600
Speaker 2: every single tree, It matters to every single industry. And

124
00:08:01,640 --> 00:08:04,120
Speaker 2: so for the very first time, there's a there's a

125
00:08:04,160 --> 00:08:08,480
Speaker 2: whole new computer market that is going to be uh

126
00:08:08,720 --> 00:08:11,720
Speaker 2: in in every single country and every single every single market.

127
00:08:12,000 --> 00:08:15,920
Speaker 2: And UH it starts with it starts with, of course,

128
00:08:16,480 --> 00:08:22,200
Speaker 2: UH the native computer industry itself. But you're seeing you're

129
00:08:22,240 --> 00:08:27,360
Speaker 2: seeing a great adoption in healthcare, great adoption and logistics Uh.

130
00:08:27,720 --> 00:08:31,680
Speaker 2: In in transportation, of course, UH. In manufacturing, in the

131
00:08:31,760 --> 00:08:35,400
Speaker 2: large industries, the heavy industries. Uh. For the very first time.

132
00:08:36,040 --> 00:08:39,760
Speaker 2: Because of generative AI, computers are going to be computer

133
00:08:39,880 --> 00:08:43,360
Speaker 2: technology is going to impact literally every single industry and

134
00:08:43,440 --> 00:08:45,840
Speaker 2: every single country, and so so the markets are going

135
00:08:45,880 --> 00:08:46,600
Speaker 2: to be quite large and.

136
00:08:46,520 --> 00:08:50,320
Speaker 1: Global, Jensen. Our final question comes from our audience. Actually,

137
00:08:50,360 --> 00:08:52,360
Speaker 1: I said that you were coming on and I think

138
00:08:52,400 --> 00:08:55,360
Speaker 1: you'll appreciate this one, and it relates to sovereign AI.

139
00:08:55,920 --> 00:08:58,080
Speaker 1: But it but it's how AI impacts all of us.

140
00:08:58,120 --> 00:09:01,839
Speaker 1: And this user asks Ray, as was prediction that human

141
00:09:01,960 --> 00:09:04,920
Speaker 1: level intelligence in AI will be achieved by the end

142
00:09:04,920 --> 00:09:09,360
Speaker 1: of this decade. Does Jensen believe that that trajectory and

143
00:09:09,440 --> 00:09:10,600
Speaker 1: timeline is on track?

144
00:09:13,200 --> 00:09:16,960
Speaker 2: Well, uh, you know, a g I from an engineer's perspective,

145
00:09:17,160 --> 00:09:23,680
Speaker 2: requires requires specification. And if if a g I specification

146
00:09:24,200 --> 00:09:30,080
Speaker 2: was a collection of tests that humans are able to

147
00:09:30,160 --> 00:09:33,960
Speaker 2: perform and until now no computers can, and that that

148
00:09:34,120 --> 00:09:38,560
Speaker 2: suite of tests, whether it's math or English, or whether

149
00:09:38,600 --> 00:09:42,080
Speaker 2: it's you know, it could be law tests, or it

150
00:09:42,080 --> 00:09:46,080
Speaker 2: could be uh, you know, medical, whatever, whatever the tests are,

151
00:09:46,200 --> 00:09:48,959
Speaker 2: if that suite of tests could be specified, could be

152
00:09:48,960 --> 00:09:52,079
Speaker 2: put in front of generative AI over the course of

153
00:09:52,360 --> 00:09:56,800
Speaker 2: of the next you know, within this decade. My my

154
00:09:56,960 --> 00:10:00,120
Speaker 2: guess would be that at the rate of the current progress,

155
00:10:00,400 --> 00:10:02,680
Speaker 2: it is very likely that that suite of tests would

156
00:10:02,679 --> 00:10:09,760
Speaker 2: be achieved by a computer. However, they're much larger definitions

157
00:10:09,800 --> 00:10:15,080
Speaker 2: of human general intelligence, and until we're able to specify

158
00:10:15,080 --> 00:10:17,680
Speaker 2: what that means, or even understand what that means, it's

159
00:10:17,720 --> 00:10:19,160
Speaker 2: going to be very hard to know whether we have

160
00:10:19,160 --> 00:10:23,360
Speaker 2: achieved it or not. But the definition that I provided,

161
00:10:23,720 --> 00:10:28,360
Speaker 2: which is the ability for a computer to achieve excellent

162
00:10:28,360 --> 00:10:33,040
Speaker 2: results on a suite of tests that previously will given

163
00:10:33,040 --> 00:10:36,000
Speaker 2: to human that suite of tests I think within the

164
00:10:36,040 --> 00:10:39,560
Speaker 2: decade will be achieved by computers in general. Today I

165
00:10:39,640 --> 00:10:42,240
Speaker 2: will be would be a tool that could be used

166
00:10:42,360 --> 00:10:46,240
Speaker 2: in a large field of science and many fields of

167
00:10:47,200 --> 00:10:47,840
Speaker 2: the industries.

168
00:10:47,880 --> 00:10:50,840
Speaker 1: And Gensen one